UvA-DARE is a service provided by the library of the University of Amsterdam (https://dare.uva.nl)
Dutch children’s acquisition of verbal and adjectival inflection
Polišenská, D.
Publication date
2010
Link to publication
Citation for published version (APA):
Polišenská, D. (2010). Dutch children’s acquisition of verbal and adjectival inflection. LOT.
General rights
It is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s), other than for strictly personal, individual use, unless the work is under an open content license (like Creative Commons).
Disclaimer/Complaints regulations
If you believe that digital publication of certain material infringes any of your rights or (privacy) interests, please let the Library know, stating your reasons. In case of a legitimate complaint, the Library will make the material inaccessible and/or remove it from the website. Please Ask the Library: https://uba.uva.nl/en/contact, or a letter to: Library of the University of Amsterdam, Secretariat, Singel 425, 1012 WP Amsterdam, The Netherlands. You will be contacted as soon as possible.
4
Production of finite verbal inflection
This chapter presents an empirical investigation of Dutch children’s acquisition of finite verbal inflection. The aim of this investigation is to address the research questions presented in Chapter 3 by analyzing elicited production data. The focus of the investigation is twofold. First, in order to test the claim that children have early knowledge of inflection, I examine the extent to which the rules for finite verbal inflection are productive in monolingual Dutch children. Second, I explore whether or not the order of acquisition of verbal forms can be accounted for by considering the degree of salience of each particular morpheme. Recall from Chapter 3 that, based on salience, one would not expect variation between the rates of acquisition for the finite morphemes. This chapter is organized as follows: In Sections 4.1 through 4.5, I provide details on the methodological design of the study. Results are presented in Section 4.6, followed by an interpretation of results in 4.7. Section 4.8 concludes the chapter.
4.1
Participants
In total, 46 monolingual Dutch children aged three to six years participated in the study. The children lived in the central, western part of the Netherlands and spoke the standard variety of Dutch. There are regional dialects of Dutch which show variation in their inflectional systems (Bennis and MacLean, 2006; Aalberse, 2009; MacLean, in preparation). It was therefore necessary to ensure that all children were acquiring the same variety of Dutch.
Since the aim of the study is to provide insight into Dutch children’s typical development, it was necessary to exclude children with developmental problems that could influence language development. According the children’s teachers, none of the selected children had experienced any developmental
problems with cognitive skills, language skills, or socio-emotional skills. Nor, did any of the children suffer from hearing loss or vision problems. Children who had attended speech therapy were also excluded from this study.
Dutch children’s productive use of inflection starts around two and half and continues to develop beyond three years (De Haan, 1996; Blom, 2003). More specifically, De Haan reports that, from two and a half years till three years, children become less accurate with their use of finite suffixes. I begin at the age where De Haan’s work ended, that is, with three-year-olds. Choosing children at this age will also ensure that children are capable of participating in the elicitation tasks. Based on the findings from cross-linguistic research (Bittner et al., 2003), I assume that verbal inflection should be mastered by six years. Accordingly, this is the age at which I stop collecting data.
I used a cross-sectional design according to which children were divided into four age groups. Age is thus the independent variable, while knowledge of inflection is considered the dependent variable. Table 4.1 presents information about the number of participants in each age group, their sex, and the means and standard deviations of their age.
Table 4.1: Information about the participants in the verbal inflection test
AGE
GROUP N OF PARTICIPANTS MEAN AGE SD AGE
(YRS) BOYS GIRLS TOTAL (YR; MOS)
3 6 6 12 3;07 0.24
4 5 7 12 4;05 0.41
5 5 7 12 5;05 0.32
In order to compose a representative sample, I selected children who came from parents with low, middle, and high socio-economic statuses (SES). In general, however, most of the children had middle to high SES backgrounds. Although SES has been reported to influence children’s developing vocabulary skills (e.g. Horton-Ikard and Ellis Weismer, 2007) and their sentence length (e.g. Snow, 1999), there has been, to my knowledge, no reported influence of SES on the development of morphosyntactic agreement (e.g. Chiat, 2001).
4.2
Data collection
Parents always granted consent to allow their children to participate in the study. In order to guarantee children’s anonymity, all names have been changed; only information relating about age and gender is documented. The data were collected in a controlled experimental setting. The children were tested in a separate room at their local school or in their day care centre. The testing sessions were audio-recorded by a digital recorder (Olympus HD-10) that was equipped by a microphone and placed in the middle of the table. Children’s responses were orthographically transcribed following the spelling conventions of Standard Dutch (see Appendix 4.1 for a scoring sheet). The transcriber involved in this project was a graduate student who received detailed instruction as described in coding book FlexiT (Blom, Orgassa and Polišenská, 2008). The transcriber was not related in any way with any of the participants. In order to assess reliability, 20% of the recordings (randomly selected) were transcribed by a second transcriber. Agreement was very high, at 93%.
4.3
Test conditions and test items
It was also crucial to collect information about children’s knowledge of the complete set of paradigmatic contrasts. This was achieved by testing different forms individually. In total, I elicited five forms, all of which are listed in Table 4.2.
Table 4.2: An overview of the test conditions
TEST CONDITION STEM + SUFFIX EXAMPLE (LEZEN ‘TO READ’)
1SG STEM + ø Ik lees
2SG-INV STEM + ø Lees jij
2SG STEM + t Jij leest
3SG STEM + t Hij leest
3PL STEM + en Zij lezen
I tested children’s knowledge of finite verb inflection in non-inverted order (subject-verb) in 1SG,2SG,3SG and 3PL. For the 2SG condition, I also tested the
inverted order (verb-subject). This decision was made because the Dutch verbal paradigm includes a contrast in 2SG between the non-inverted and the inverted order. Namely, the inverted finite verb in 2SG is bare whereas the non-inverted
finite verb in 2SG is marked by the –t suffix. Examples of this contrast are provided in (1a) and (1b).
(1) a. lees jij
read –2SG-INV you
‘you read’
b. jij leest
you read –2SG
‘you read’
All conditions were tested with both existing and nonce verbs. The inclusion of nonce verbs is essential because it provides information on how productive children really are in applying inflectional rules (see also Chapter 1). Children’s correct use of a correct existing finite verb can either reflect their knowledge of
inflectional rules, or it can reflect their storage of unanalyzed forms (Berko, 1958; Peters, 1982; Pinker, 1999; Tomasello, 2003). If children use inflection productively, we expect good performance with nonce verbs since nonce verbs cannot already be stored on the basis of input data. The main criterion for the choice of the nonce verbs was that they would be unknown to children.
In order to ensure that children were actually familiar with the existing verbs, test words were selected from the standardized vocabulary list for Dutch children under the age of three (N-CDI, Zink and Lejaegere, 2002). The existing verbs denoted actions that could be easily modeled by both the researcher and the child. Table 4.3 presents the test items.
Table 4.3: An overview of the items in the verbal inflection test
EXISTING VERBS NONCE VERBS
Tekenen ‘to draw’ Pieren
Drinken ‘to drink’ Spollen
On the basis of the five conditions and four test items, a maximum of 20 responses per participant were possible. A number of children were tested with an earlier version of the test that contained three existing items and three nonce items. This, of course, resulted in a maximum of 30 responses per participant. In total, 17 children were tested with the earlier version: five three-year-olds, two five-year-olds, and ten six-year-olds. The additional nonce item was kluiken ‘to kluik’ and the additional existing item was trekken ‘to pull’. The reason to (slightly) adapt the test during the study and limit the number of test items to four was that elicitation of six verbs turned out to be too demanding on younger children’s attention abilities. Consequently, during the testing, children would indicate that they no longer wanted to take part in the session, often giving answers such as ik weet het niet ‘I don’t know’. Given that the main goal of the test was to elicit various finite morphemes, I preferred that the child finish all tasks, and produced the complete set of finite morphemes with less test items. Table 4.4 provides an overview of the maximum number of elicited
responses per group, taking into consideration, the different number of test items across subjects.
Table 4.4: An overview of maximum obtained responses in each group
AGE GROUP EXISTING VERBS NONCE VERBS TOTAL
3 (N=12) 145 145 290 4 (N=12) 120 120 240 5 (N=12) 130 130 260 6 (N=10) 150 150 300
4.4
Procedure and materials
The children were tested individually in a controlled experimental setting. The test session was run by two researchers. The aim of the experiment was to elicit a subject, a finite verb and a direct object. This combination is crucial for determining whether or not the child is able to appropriately use a finite verb form. Recall from Chapter 3 that the Dutch infinitive and the finite plural form are both marked by the suffix –en. If a child placed a verb with an –en suffix in front of the direct object, she was thought to use a finite form as in (2a). However, when the same verb form follows the object, the child was thought to produce a non-finite form (i.e. root infinitive) as in (2b).
(2) a. Kinderen tekenen een zon.
Children draw [+FINITE] a sun [OBJECT]
b. Kinderen een zon tekenen.
The test was presented to the children as a game, in order to maintain the children’s attention. The session started by introducing the nonce verbs, after which the child took part in two tasks: a sentence completion task, and an activity task. For the sentence completion task, I used colored printed photographs which were taken with a digital camera. The photographs were placed in a booklet in a pseudo-randomized order. Filler pictures which were not relevant to children’s knowledge of verbal inflection were also included. The fillers represented items from the gender attribution test, which is discussed in detail in Chapter 6. Table 4.5. provides an overview of the objects that I used in the activity tasks.
Table 4.5: An overview of objects used for performing the activities
TEST ITEM OBJECTS
Drinken ‘to drink’ a cup
Tekenen ‘to draw’ paper and a pencil
Pieren ‘to pier’ a pier* and a spoon
Spollen ‘to spol’ a spoller* and a toothbrush
*Images of these objects are given in (3).
Because young children are often more willing to interact with a puppet than they are with an adult (Thornton, 1996), a puppet was used in two of the activity tasks. The puppet appeared in the test session after the introduction of the nonce verbs. He then asked the child if she would participate in a game. The experimenter told the child that the puppet was too shy to talk to grown-ups but was willing to talk to children. The experimenter explained that the child’s help was very important because otherwise the game could not continue. Table 4.6 shows how the test session was structured, and offers additional information about whether the task included a puppet and/or assistance of a second researcher.
Table 4.6: A scheme of a test session
TEST SCHEME PUPPET ASSISTANCE
Introduction nonce verbs no no
Introduction puppet yes yes
Activity description task 2SG-INV yes yes
Activity description task 1SG yes yes
Activity description task 2SG no yes
Sentence completion task
3SG/3PL no no
Four of the six parts in the test session required assistance of a second experimenter. The second experimenter either played the role of the puppet or assisted the child by handing her the objects. In the remaining conditions, the assistant observed the session and took notes. In total, the test lasted approximately 30 minutes. In general, the younger the child, the longer the test session took. Each task included a preliminary test trial to make sure that the child understood the instructions correctly. The verb lezen ‘to read’ was used as a practice item. In the paragraphs below, I will explain how nonce verbs were introduced and how the tasks among the various conditions were carried out.
Introduction nonce verbs
The experimenter used two novel objects named a ‘pier’ and a ‘spoller’ in order to introduce the nonce verbs. (3) provides images of pier and spoller.
(3) Images of a ‘pier’ (on the left) and a ‘spoller’ (on the right)
To avoid triggering infinitival forms or imitation of finite forms in the present tense, the nonce verbs were presented in the past participle form. An example of the nonce verb pieren can be found in (4), along with its other morphological forms.
(4) a. infinitive: pieren
b. finite forms: pier, piert, pieren
c. past participle: gepierd
During the introductory session, the experimenter encouraged the child, herself to act out and name the activities. After introducing the nonce verbs, the experimenter verified that the child had memorized them accurately. Elicitation tasks only began after experimenters were sure that children had accurately learned the nonce verbs. If, during the elicitation task, the child failed to retrieve the nonce verb, the experimenter was permitted to help the child by naming the object. If this cue did not help, the introduction session was repeated. The procedure for introducing nonce verbs is given in (5). The italics represent the Dutch forms, in which the nonce verbs are introduced.
(5) Now we are going to do something funny. Look, I’ve brought some things with me. This is a ‘spoller’ and that is a ‘pier’ [experimenter shows the objects]. Look what you can do with it [experimenter performs the
activity]. With the spoller, I have gespold and with the pier I have gepierd. Now it is your turn.” [experimenter and the child play together with the objects for several minutes until the child can name the activities herself]
During the introduction session, it was apparent that some three-year-old children were not at ease in learning the new words. This resulted in a situation where children did not want to repeat the nonce verbs or stopped talking altogether. In these cases, the experimenter ended the session. In total, five sessions were ended after the introduction.
Activity task for 2SG-INV
The aim of this task was to elicit finite verbs in 2SG-INV in inversion. This was achieved by letting the child ask a WH-question to the puppet. The task proceeded as follows: While the puppet was performing an activity with the object(s), the experimenter instructed the child to ask the puppet a WH
-question. An example is given in (6), which illustrates the task for the verb
tekenen ‘to draw’.
(6) PUP: [performs the activity of drawing]
EXP: [instructs the child to ask the puppet WH-question]
De pop is aan het tekenen. Vraag hem wat.
‘The puppet is drawing. Ask him what.’ CHI: Wat teken je?
‘What are you drawing?’ PUP: [gives a response to the child]
Although the instruction of this task was generally simple for children to understand, some three-year-olds had difficulty forming WH-questions. In these cases, children either provided the answers themselves, or repeated the WH
-word without adding the verb. In total, two children failed to respond on this task.
Activity task in 1SG
The aim of this task was to elicit finite verbs in the 1SG. At the beginning of
explained to the child that the puppet was very curious about what was happening, but, unfortunately, was not able to see it. An example of the task procedure for a nonce item pieren is given in (7).
(7) EXP: [gives a ‘pier’ and a spoon to the child]
CHI: [takes the objects and performs the novel activity] PUP: Wat gebeurt er?
‘What’s happening?’ CHI: Ik pier een lepel
‘I pier (nonce verb) a spoon.’ PUP: [gives a response to the child]
In this task, some three- and four-year-olds had difficulty taking the perspective of a blindfolded puppet and hence, did not see why they would have to describe the ongoing action. In order to prevent loss of responses, it was helpful to first let the child experience the blindfold for themselves.
Activity task in 2SG
The aim of this task was to elicit finite verbs in 2SG. At the beginning of this
task, one experimenter was blindfolded. The blindfolded experimenter also performed the activities. The task of the child was to describe to the blindfolded experimenter, the ongoing activity since she could not see it herself. The second experimenter assisted the child by handing over the objects. In (8), is an example of the task for an item drinken ‘to drink’.
(8) CHI: [gives a cup to the blindfolded experimenter] EXP: [takes the cup and pretends drinking]
Wat gebeurt er?
‘What’s happening?’ CHI: Je drinkt uit een beker
‘You’re drinking from a cup.’ EXP: [gives a response to the child]
In this task, some children correctly noted that, even if the experimenter could not see the object(s), she was able to feel what she was doing. Thus, for these children, it did not seem logical to describe the ongoing activity to the blindfolded experimenter. However, in this task, the loss of responses was
minimal, as children generally had no problem accepting the explanation that it was simply part of the game.
Sentence completion task in 3SG and 3PL
The sentence completion task was used to obtain data about the 3SG and 3PL
contexts in declarative main clauses. In this task, contrasting images were presented in pairs and children were instructed to describe the contrast between the two adjacent pictures, e.g. reading a book vs. reading a newspaper by completing the sentence which was started by the experimenter. This is illustrated in (9).
(9) Example of elicitation of the 3SG and 3PL verb forms by means of sentence
completion task.
dit meisje leest een boek
‘this girl is reading a book’
en de ouders lezen een krant
‘and the parents are reading a paper’ This task proceeded as follows: The experimenter triggered the sentence by pronouncing the underlined words in (9) and the participant’s role was to complete the sentence (the correct subject responses are in bold). It should be mentioned that a potential risk of this task was that children might single out one of the characters when describing the action depicted in the 3PL test items.
Consequently, when a child produced 3SG instead of 3PL, it was not possible to resolve whether that child referred to only one of the two characters and
correctly produced the 3SG or whether she produced incorrect inflection. In order to avoid this ambiguity as much as possible, I only chose subjects that were naturally plural such as de kinderen ‘the children’ or de ouders ‘the parents’. In addition, the experimenter introduced each picture by naming the subjects first. For example, in (9), the experimenter would first point at the pictures and say ‘Look, here is a girl and here are her parents’. After this introductory sentence, the experimenter proceeded with the sentence completion task as described above.
4.5
Data analysis
Children’s responses were divided into five categories: (1) correct finite verb, (2) incorrect finite verb, (3) alternative construction, (4) irrelevant response and (5) unintelligible or no response. A finite verb was scored as ‘correct’ when a target morpheme was realized in a given context. Contrastively, a finite verb was scored as incorrect when a non-target morpheme was produced in a given context. An example of a correct response in the second person is provided in (10a) and examples of incorrect responses are provided in (10b).
(10) a. jij tekent een zon
‘you draw a sun’ b. jij teken een zon
jij tekenen een zon
The category of alternative constructions refers to responses in which the lexical verb (i.e. the test item) was not finite. Instead, the lexical verb was produced in its non-finite form in sentence final position. The following constructions were also scored as alternative constructions: the present progressive in (11), root infinitives in (12), and so called dummy auxiliaries (i.e. periphrastic verbs that consist of auxiliary + infinitive (Jordens, 1990; Hollebrandse and Roeper, 1996; Van Kampen, 1997; Zuckerman, 2001)) as illustrated in (13).
(11) Ik ben water aan het drinken.
I am water drink –INF
(12) water drinken
water drink –INF
‘drink water’ (13) ik ga water drinken
I go water drink –INF
The response was scored as a root infinitive when the verb containing the –en inflection followed the direct object and no auxiliary was present (see also Section 4.4). Note that dummy auxiliaries, such as in (13), do not contribute to the utterance’s meaning; that is, a sentence with dummy auxiliary + infinitive has the same denotation as a sentence with a finite main verb (Zuckerman, 2001). Dummy auxiliaries are present in order to make the sentence finite (Jordens, 1990). The category of dummy auxiliaries includes the verbs gaan ‘go’ and doen ‘do’. Whereas gaan ‘go’ in adult Dutch denotes a future meaning, children may use it to denote an ongoing activity. For example, during a test session a child might say Ik ga tekenen ‘I go draw’ although she has already started drawing. As for the verb doen ‘do’, a construction such as Jij doet drinken ‘You do drink’ is considered appropriate only in Southern dialects in the Netherlands (Cornips, 2000; Blom and DeKorte, 2008).
The category of irrelevant responses includes omissions of a direct object as in (14a) and uses of a non-target verb as in (14b).
(14) a. de kinderen lezen
‘the children read’ b. de ouders staan bij de muur
‘the parents stand near the wall’
4.6 Results
In order to keep the overview of the results clear, this section is divided into four subsections. In 4.6.1, I will describe how the response types are distributed across existing and nonce verbs. In 4.6.2, I will report on data quantification by means of percentages of correct inflection, whereas in 4.6.3, I will report on effects of age and test conditions on children’s linguistic productions. Subsection 4.6.4 focuses on error analysis.
4.6.1
Response types
In order to find out whether or not children performed differently with existing and nonce verbs, I will begin with an overview of the distributions of response types, as presented in Tables 4.7 and 4.8.
Table 4.7: Response types with existing verbs
Age
group inflection Correct inflection Incorrect construction Alternative Irrelevant Unintelligible/No response
3 (N=145) 65 % 4 % 24 % 3,5% 3,5% 4 (N=120) 85 % 6 % 6 % 1,5% 1,5% 5 (N=130) 92 % 2 % 4 % 0,5% 0,5% 6 (N=150) 98 % 0 % 2 % 0 % 0 %
N=number of analyzed responses
Table 4.8: Response types with nonce verbs
Age
group inflection Correct inflection Incorrect construction Irrelevant Alternative Unintelligible/No response
3 (N=145) 64 % 7 % 23 % 3 % 3 % 4 (N=120) 82,5% 5 % 7,5% 2,5% 2,5% 5 (N=130) 94 % 3 % 2 % 0,5% 0 % 6 (N=150) 90,5% 0 % 8 % 0,5% 0,5%
The percentages presented in Tables 4.7 and 4.8 indicate that children’s response types were, more or less, the same with nonce verbs and existing verbs (for information about the raw figures, see Appendix 4.2). The overall percentages of incorrect inflection (all conditions collapsed) were extremely low across age groups (between 0 – 7%). It is evident that three-year-olds produced a higher percentage of alternative constructions as compared to other groups. It is possible that children who used alternative constructions where the target verb is in an infinitival form, were actually avoiding the finite verb inflection. I will return to this issue in Section 4.7.
Correlational analyses were carried out in order to compare children’s within-subject performance on nonce versus existing verbs. Kendall’s non-parametric test of correlation revealed a significant association between the frequency of children’s correct inflection with existing and nonce verbs (τ = .77, p < .01). Similarly, a significant correlation was found between children’s use of alternative constructions with existing and nonce verbs (τ = .835, p < .01). The results of the correlations demonstrate that children’s performance was consistent with existing verbs and nonce verbs.
Returning to Tables 4.7 and 4.8, unanalyzable responses (irrelevant responses, unintelligible responses, and no responses) fell under 5% in each age group. These responses are not included in further analyses.
4.6.2
Accuracy
The aim of the accuracy analysis was to assess the extent to which children at different ages produced correct finite verbal inflection. This analysis focused solely on subject – verb – object responses and on wh-word – verb – subject responses since these constructions are the only ones which allow us to conclusively evaluate the degree to which children correctly use finite verbal inflection (see also Section 4.4). It is worth mentioning that there is no pre-established criterion for establishing whether an item is or is not acquired. Some studies look at the first correct appearance of a given morpheme whereas others establish a cut-off point based on children’s percentage of correct usage in obligatory contexts. Brown’s (1973) criterion is one of the better known criteria, which states that a morpheme is considered as acquired when it appears correctly in at least 90% of obligatory contexts (Brown, 1973). It should be noted, however, that Brown set the cut-off point for acquisition at 90% because he observed that, when children’s acquisition data were represented
graphically, the accuracy curve for some particular morphemes was initially chaotic. Once that curve had passed the 90% line, however, it tended to remain above that line. This observation implies that it is not necessarily the 90% threshold that guarantees the acquisition of a morpheme since the criterion is somewhat arbitrary. Instead, a more reliable assessment would entail determining when children are able to use a particular morpheme systematically and consistently with a high accuracy rate. In other words, it is reasonable to believe that children might actually have acquired a particular morpheme, without having reached Brown’s 90% mark, assuming they are consistently accurate for an extended period of time.
Tables 4.9 and 4.10 show children’s accuracy rates for existing verbs and nonce verbs. The test conditions are listed in the first row in the tables. Table 4.9 shows that children in all age groups are highly accurate with existing verbs.
Table 4.9: Existing verbs: % correct in various test conditions
AGE GROUP 1SG 2SG 2SG-INV 3SG 3PL 3 (N=12) 94% 15/16 95% 19/20 93% 25/27 88% 15/17 100% 14/14 4 (N=12) 100% 19/19 81% 17/21 95% 19/20 89% 17/19 100% 23/23 5 (N=12) 100% 25/25 92% 23/25 100% 26/26 95% 21/22 100% 22/22 6 (N=10) 100% 30/30 100% 30/30 100% 30/30 100% 29/29 100% 28/28
However, one may notice that the four-year-olds are accurate only 81% of the time in the 2SG condition. This is relatively low compared to the other
conditions. In total, the four-year-olds produced four inflectional errors out of 21 scorable responses. However, two errors in this condition can be attributed to a single child who persistently did not produce the –t suffix in the test. This particular child replaced the target –t suffix with either the –ø suffix or used an
alternative construction. In the conditions which did not require the –t suffix, the child’s performance was accurate. The remaining two errors were produced by two different children.
Table 4.10 presents the results of the accuracy analysis with nonce verbs.
Table 4.10: Nonce verbs: % correct in various test conditions
AGE GROUP 1SG 2SG 2SG-INV 3SG 3PL 3 (N=12) 72% 13/18 100% 20/20 85% 22/26 100% 14/14 93% 14/15 4 (N=12) 100% 16/16 94% 17/18 100% 22/22 82% 18/22 95% 20/21 5 (N=12) 100% 25/25 92% 23/25 96% 25/26 100% 23/23 95% 22/23 6 (N=10) 100% 26/26 100% 27/27 100% 30/30 100% 25/25 100% 28/28
Children across age groups were highly accurate with nonce verbs. In most conditions, the accuracy level was above 90%. In this respect, the children’s performance with nonce verbs resembled their performance with existing verbs. A percentage lower than 90% was found in two groups: the group of three-year-olds in the 1SG condition (72%; five errors); and in the 2SG-INV
condition (85%; four errors). Each of these conditions required the –ø suffix. The accuracy rate for the four-year-olds was 82% in the 3SG condition. This
figure reflects the performance of a single child (who was already discussed with regard to the accuracy in existing verbs), who consistently failed to produce the –t suffix. This child was responsible for two of the four errors. One might expect that this particular child’s performance would also affect the accuracy of the 2SG in the four-year-olds since this condition also requires the –
t suffix. With the data currently available, this possibility cannot be ruled out
since, in each 2SG condition, this child systematically produced alternative constructions instead of finite verbs.
The performance of the child in the 4-year-old group who consistently failed to produce the –t suffix could mean that many of the errors were produced by a small number of children. This is not the case, however. Instead, inflectional errors are distributed evenly across children and cannot be ascribed to one or two individual subjects. In the younger groups (the three- and four-year-olds), at least half of the children err at least once in the test.
4.6.3
Effect of test conditions and age
The accuracy analysis in the previous section showed that children produced correct inflection far above chance level for both existing and nonce verbs. However, the accuracy analysis in 4.6.2 did not take into account the relationships between the three variables: age, test condition, and accuracy rate. A more robust statistical test was therefore needed to investigate the extent to which one variable can be explained or predicted by one or more of the other variables. For this data, a Multinomial Logistic Regression (henceforth: MLR) was appropriate (as described in Nooteboom and Quené, 2008). The aim of MLR was to address two questions. The first question was whether or not children’s accuracy-rates varied per condition and the second questions was whether children’s accuracy-rates varied across ages. Before I turn to the presentation of the results I will briefly address the basic principles of the MLR analysis.
MLR is appropriate to use when one is interested in determining the impact of more than one independent variables on a dependent variable. The dependent variable must be nominal and must consist of more than two categories.6 In the present study, the independent variables are age and test
condition; and the dependent variable is children’s response on the verbal inflection test. The dependent variable is nominal and consists of a set of three categories: (1) correct inflection, (2) incorrect inflection and (3) alternative construction. When using MLR, one category of the dependent variable must be chosen as the comparison category in order to determine a proportion. For this data, the category of correct inflection was chosen as the comparison category. The MLR determined whether the proportions of incorrect responses
6 Nominal variable refers to a set of categories, which cannot be ordered in any meaningful way.
In contrast, an ordinal variable refers to a set of categories, which can be ordered on a scale such as highest degree or social class. In addition, when the nominal variable is dichotomous in nature (there are only two categories) the statistical analysis known as logistic regression is used.
and the proportions of the alternative constructions were affected by the independent variables (test conditions and age).
My data contained two kinds of variation with respect to children’s responses: First, I did not obtain an equal number of scorable responses from each child and second, the individual children varied in their response category within test conditions. An example of the latter would be if, in the 1SG, a child
responded differently to each of the four test items. In an extreme example, a child could have, for example used correct inflection with the first test item, incorrect inflection with the second, an alternative construction with the third, and no response with the forth. In order to assess the variability in the data, I used a two stage-bootstrap with replacement procedure (Nooteboom and Quené, 2008). The bootstrap technique works as follows: A random sample of 44 children out of 45 was drawn with replacement. From the 44 children, 15 responses are drawn with replacement. This procedure is repeated 250 times, which provides 250 regression coefficients. The distribution of the coefficients indicates the variability across the participants as well as the variability across the responses.
Tables 4.11 and 4.12 summarize the mean coefficients, the floor (2.5%) and the ceiling (97.5%) level of the confidence intervals for incorrect inflection in existing and nonce verbs, respectively.
Table 4.11: Coefficients for incorrect inflection in existing verbs.
COEFFICIENTS VARIABLE MEAN 2.5% 97.5% 1SG -12.80 -59.03 -2.93 2SG -3.19 -12.98 -1.18 2SG-INV -6.58 -22.22 -2.43 3SG -3.97 -16.03 -1.68 3PL -19.33 -77.76 -10.45 AGE -1.08 -2.55 -0.13
Table 4.12: Coefficients for incorrect inflection in nonce verbs COEFFICIENTS VARIABLE MEAN 2.5% 97.5% 1SG -3.80 -16.41 -1.47 2SG -4.36 -16.36 -1.86 2SG-INV -4.12 -12.43 -2.09 3SG -4.58 -17.77 -1.83 3PL -4.42 -17.21 -1.64 AGE -1.04 -1.99 -0.33
The negative values of the coefficients imply that children’s produce incorrect inflection less frequently than they do correct inflection The mean coefficients with the highest negative value, such as in 1SG ad 3PL in Table 4.11, show that children were almost error-free in these conditions. However, the overlap of the confidence intervals across various conditions means that none of the conditions significantly affected children’s realization of incorrect inflection. In other words, none of the conditions posed a greater problem for children than the others. With regard to effects of age, the crucial information was whether the ceiling level (97.5%) of the confidence interval differs from the zero value. As can be seen in both Table 4.11 and Table 4.12 the 97.5% confidence interval indicates a negative value for age. This implies that one can be 97.5% confident that the proportion of incorrect inflection decreased with age.
Similar results were found for children’s production of alternative constructions. Tables 4.13 and 4.14 presents the mean coefficients, the floor and ceiling confidence intervals for the proportion of correct inflection and alternative constructions in existing and nonce verbs, respectively.
Table 4.13: Coefficients for alternative constructions in existing verbs. COEFFICIENTS VARIABLE MEAN 2.5% 97.5% 1SG -2.09 -3.49 -1.13 2SG -2.67 -4.40 -1.25 2SG-INV -9.08 -42.80 -2.8 3SG -1.68 -3.12 -0.78 3PL -1.76 -3.39 -0.65 AGE -1.28 -2.57 -0.58
Table 4.14: Coefficients for alternative constructions in nonce verbs.
COEFFICIENTS VARIABLE MEAN 2.5% 97.5% 1SG -1.45 -2.81 -0.49 2SG -2.47 -4.18 -1.33 2SG-INV -8.34 -22.82 -2.55 3SG -1.41 -2.52 -0.58 3PL -1.68 -2.66 -0.77 AGE -0.9 -1.87 -0.18
The overlap of confidence intervals in Tables 4.13 and 4.14 indicates that the proportion of alternative constructions did not vary significantly across the test conditions. As was the case with the incorrect forms, a significant negative effect of age was observed for both existing and nonce verbs, indicating that the use of alternative constructions decreases with age.
4.6.4
Error profile
Sections 4.6.2 and 4.6.3 have shown, first of all, that the children were highly accurate in using inflection, both with existing and nonce verbs. The accuracy analysis revealed that the amount of inflectional errors that children produced was 4% (36 errors in 913 finite contexts). However, how systematic are the errors that do occur? This question is relevant to the discussion of form-feature specification of the Dutch verbal system because a particular feature specification may prevent children from using a certain morpheme, which could inflate the use of another (underspecified) morpheme.
The results in 4.6.3 address this issue by showing that children’s error-rates did not vary significantly across conditions. If the children showed a clear-cut error profile, e.g. if they had overused only one particular morpheme, this would have resulted in significant differences across conditions. For example, a strong preference for the –ø suffix would have resulted in a relatively high proportion correct in 1SG as well as in 2SG-INV and less accurate behavior in
other conditions. If the children had preferred the –t suffix, this would have resulted in a high proportion correct in the 2SG and 3SG and lower
accuracy-rates in all the rest. Finally, if the children had preferred the –en suffix, the plural condition would have showed higher accuracy than singular conditions. However, because of the low number of errors, such a quantitative approach did not work. Therefore, in this section I will take a closer look at children’s error profiles. Table 4.15 shows the distribution of the morphemes across conditions. The grey cells in the table represent the contexts in which the morpheme was realized correctly. The error-analysis is based on data of children between three and five years. I excluded data from the six-year-olds because they reached 100% accuracy in all conditions.
Table 4.15: Error-profiles in verbal inflection
CONTEXT STEM + ø STEM + t STEM +en
1SG 95 % 113/119 5 % 6/119 0 % 0/119
2SG 7,5% 10/129 92,5% 119/129 0 % 0/129
2SG-INV 94,5% 139/147 5,5% 8/147 0 % 0/147 3SG 7,5% 9/117 92,5% 108/117 0 % 0/117
3PL 1 % 1/118 1,5% 2/118 97,5% 115/118
Table 4.15 shows a striking contrast between the –en suffix, on the one hand, and the –t and the –ø suffixes, on the other. Whereas no instances of overuse of the –en suffix were found in finite contexts, overuse of both the –t and the –ø suffixes were present. In other words, when children used a finite verb inaccurately, they produced the –t or the –ø suffix. Overuse of –en was not attested.
4.7
Interpretation of the results
The goal of the present investigation was to gather new evidence in order to examine the acquisition of finite verbal morphemes by analyzing elicited production data from Dutch monolingual children between ages three and six. The questions addressed in this study were: (1) To what extent do the data from child
Dutch support the idea that children know agreement inflection from early on? And (2) To what extent can salience account for the order of acquisition of inflectional morphemes in monolingual Dutch children?
The main finding is that finite verbal inflection is acquired at the age of three. By carefully analyzing the data for five inflectional contexts, I found high percentages of accuracy in both existing and nonce verbs. In Section 4.3, I
argued that performance with nonce verbs reflects productivity of inflection rules. Children’s performance with nonce verbs was therefore crucial in assessing the claim that children have early knowledge of inflection (Wexler, 1998). First of all, individual children showed consistent linguistic behavior with existing and nonce verbs: The occurrence of inflectional errors as well as occurrence of alternative constructions was highly similar for both types of verbs. This suggests that the children did have knowledge of inflection, since they are equally productive with both types of verbs. If children did not have knowledge of verbal inflection, and hence, relied on lexical storage, one would have expected to find differences in their performance with nonce verbs and existing verbs. If this were the case, children would have been expected to correctly inflect the existing verbs but err or use alternative constructions with nonce verbs, which was not the case.
Results from the accuracy analysis (Table 4.10) indicated that three-year-old children correctly used subject verb agreement with nonce verbs above 90% of the time in three of the five conditions. The few errors that the three-year-olds did make involved the 1st person singular (five errors; accuracy rate 72%) and
the 2nd person singular in inversion (four errors; accuracy rate 85%). Statistical
analysis using Multinomial Logistic Regression (MLR) (Section 4.6.3) revealed that children’s use of inflection was stable across all test conditions: Children did not produce significantly more inflectional errors nor did they produce more alternative constructions in any specific condition. This implies that monolingual Dutch children know all finite morphemes equally well. Based on these findings I conclude that Dutch three-year-olds have target-like knowledge of the verbal inflectional paradigm and that they understand that subject-verb agreement is obligatory.
Despite the overall low percentage of inflection errors (4%; 36 errors in 913 finite contexts), an effect of age did prove to be significant: Children become more accurate as they grow older. The next logical question to ask is: How might one explain this age effect? It has been proposed that children’s ability to produce particular (morphological) forms breaks down under certain task-related conditions (Hadley, 1998; Elin Thordardottir, 2008). According to Leadholm and Miller (1992), conversation is the least demanding discourse context since it is relatively unstructured and unplanned. In contrast, elicitation tasks entail one’s ability to process the task instructions, to memorize the instructions (and test items), to take the perspective of another person, and to
plan ones own response. Given these demands, an elicitation task is one of the most complex types of discourse required by children.
The less-developed skills of the younger children in the elicitation task were evident in the testing sessions: Although the youngest children demonstrated a general understanding of the tasks, they sometimes experienced difficulties in grasping what type of response the experimenter sought. For example, if for the target response jij tekent ‘you draw’ a child responded by using an alternative construction (e.g. jij gaat tekenen ‘you go draw’), the researcher encouraged the child to instead inflect the lexical verb by asking questions such as wat zeg je? kook ik ‘what do you say? do I cook’. I observed that this intervention sometimes triggered reactions such as ‘I’ve said it already’, which I interpreted as, ‘I don’t really understand what you want me to say’. In some cases, the experimenter’s intervention led to the child becoming aware that her response was not adequate. Typically, this made the child feel uneasy and the child often became reluctant to complete the task.
The tendency for the frequency of alternative constructions to decrease with age can be viewed as an extra argument for the task-effects: Children do not fail to produce correct inflection but they instead fail to produce the type of response required by the test. In addition, younger children tended to have more difficulties memorizing the nonce verbs, and they also had more difficulty to act out the tasks. This is not surprising, given that, at this age, children’s memory skills and their ability to take other people’s perspective into consideration are still developing (see Case, 1985; 1992; Case and Okamoto, 1996 for development of memory and Piaget, 1926 on development of egocentrism in children). Thus, the tendency for children to become more accurate with age and their tendency to use fewer alternative constructions with age does not necessarily suggest that younger children have not acquired the finite morphemes. Instead, their performance over time likely reflects their developing cognitive and socio-emotional skills.
Based on the hypothesis that children’s attention is drawn to the most salient morphemes, I predicted that, in Dutch verbal inflection, children would acquire the non-finite –en morpheme first, followed by the set of finite morphemes. The findings summarized in the literature overview in Chapter 3 provided sufficient evidence for the early acquisition of non-finite –en. With respect to the acquisition of finite morphemes, the data obtained in the present study do not allow me to draw conclusions about whether or not salience accounts for the acquisition of finite morphemes. This is due because Dutch
In Section 4.6.4, I argued that error analyses were necessary in order to gain insight into children’s form-feature specifications in the target system. Given that the error analysis was based on residual errors resulting from task-effects, I believe that, with respect to feature representation, error-profiles can only be informative when systematic patterns are detected. I could assume, for example, that, when confronted with increased processing demands (such as task effects), children may occasionally fall back on a less specified form. This, in turn, may provide an indication about which morphemes are more problematic for children in the earlier stages. A similar idea was proposed for second language learners in the Missing Surface Inflection Hypothesis (e.g. Haznedar and Schwartz, 1997; Lardiere, 1998; Prévost, 2003).
Returning to the children’s error patterns, it was clear that children did not use the –en suffix as a finite substitute in the singular contexts (Table 4.15). This pattern of usage is consistent with an error-pattern found in the spontaneous speech of four Dutch children aged 1;8 to 3;4 (Blom and Polišenská, 2006). Given the form-feature specification presented in Chapter 3, repeated in (15), the –en suffix is correctly specified for the features [+FIN;+PL].
(15) /t/ ↔ [+FIN;-SP;-PL] /en/ ↔ [+FIN;+PL]
/ø/ ↔ [+FIN]
/en/ ↔ [-FIN]
Importantly, the non-existent overuse of the –en in the finite contexts is in line with the claim that the –en suffix has two distinct feature representations, namely the [-FIN] –en and the [+FIN;+PL] –en (Haeseryn et al., 1997; Aalberse,
2009). In a feature representation where the –en suffix is completely underspecified and is considered the default (Wexler et al., 2004; Bennis and MacLean, 2006), such a pattern would not be expected.
The second conclusion from the error-analysis was that, within a finite paradigm, children substitute two suffixes: When children realize incorrect inflection, they overuse both the –t and the –ø suffixes across various contexts. Unfortunately, the unsystematic character of this pattern is not reliable evidence with respect to finite default. Based on the striking contrast between the finite –
en, on the one hand, and the finite –t and –ø, on the other, I speculate that, in
suffixes. If this speculation is true, I expect that, children will detect the contrastive pattern from early on. I will return to this in detail in Chapter 5.
4.8 Conclusion
Finite verbal inflection does not constitute problems for three-year-old monolingual Dutch children. In this chapter, I have presented evidence that three-year-old children have already mastered all finite morphemes. Dutch children’s high accuracy with existing verbs as well as with nonce verbs is consistent with the claims made by VEKI. Since I found that the children were no longer developing finite verbal inflection, my predictions based on salience cannot be tested on the basis of the elicited production data. The analysis of the residual inflectional errors, however, might point in a direction regarding which morphemes are more problematic for children in the earlier stages. The error-analysis suggested that the –en suffix is not completely underspecified: The existent overuse of the –en in the finite singular contexts suggests that the non-finite –en and the non-finite –en have distinct feature representations. Moreover, the error-analysis suggested that the finite –en morpheme behaves differently than the other finite morphemes. Further research with younger children is needed in order to investigate this suggestion. I will turn to this point in the next chapter.
